In additional analyses, we examined the same contrasts for the four stability/change groups, but we also controlled for stability/change on the other type of problem behavior (as well as T1 level of the predictor variable and sex). Thus, for example, when looking at associations of the pattern of externalizing problems with EC, impulsivity, and negative emotionality, we controlled for change in internalizing problems (coded 1 = SLs, 2 = IMs, 3 = Ds, 4 = SIs). The analogous analyses were conducted for internalizing problems controlling for the pattern for externalizing problems. We could not control for co-occurring symptoms by entering dummy codes for the four INT or EXT groups because too few children were in some cells if we crossed internalizing and externalizing group status in the multinomial regressions. New findings based on this method appear in italics in Table 7 and are discussed after the main analyses.
